Abstracto
- Identification of plant stress is fundamental for the improving crop efficiency, which is of economic and ecological interest around the globe. Passive remote sensing sensors measure emitted or reflected electromagnetic radiation, allowing the identification of present materials; this data is used for various plant analyses, including precision agriculture and stress detection. This paper uses hyperspectral signatures to classify of non-living negative factors (abiotic stress) in corn crops using hyperspectral signatures. The high dimensionality of hyperspectral data makes its processing highly complex, rendering dimensionality reduction essential. The two main approaches for reduction are feature selection and feature extraction; maintaining the physical meaning of data is important for interpretability, so feature extraction by band selection was used. Nitrogen deficiency was applied in a maize test field, collecting 337 samples in three nitrogen deficiency levels. Three unsupervised band selection methods were applied to the dataset. The reduced datasets obtained were classified using a Support Vector Machine, Random Forest, and a fully connected Neural Network. It was found that the low redundancy high information selected bands can equal or outperform the classification accuracy of the hyperspectral high dimensional dataset. This could allow the development of less complex multi-spectral equipment based on selected characteristics. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.